Re: [glob2-devel] Potential killer feature

[Martin Nyffenegger]

Very good Idea indeed,
I think you already came up with that one earlier.
I see a first issue, How do you do network plays with ghost ? There has 
to be a ghost-sync-selection procedure somehow to have the exact same 
ghosts on both machines.

I see a technical problem also. Computing F{} might be VERY time 
consuming therefore slowing down glob. Alors, computing the next order 
given t might be a bit more than O(n) depending on your implementation.
(Just for a hint, I do distance calculation with O(n^2) ).

But the whole idea seems very interesting and promising, I'm just 
wondering how we will come up with an efficient F{}....

Lord Barock

P.S. Just a thought about time consumption: Dumping s{t-1}, u{t} and 
o{t} in a compact form and analysing them offline (special dedicated 
programm). We don't have to dump the whole game, just interesting 
(Univers changing) events. That would perhaps be an interesting and 
powerfull solution.
Thus those player which are good could spend their time dumping and 
improving theyr ghost to make them available for othet player to 
compete against.

Le 22 déc. 04, à 15:26, Stephane Magnenat a écrit :

> Hello,
>
> I'm now in holidays. Heavy socio-politico-administrative constraints 
> from
> scientific establishment are now far away so I have time to think 
> freely.
>
> I suggest a new feature relative to AI that, if implemented correctly 
> and
> working, can really be a killer one:
> http://www.ysagoon.com/twiki/bin/view/Glob2/GhostProposal
>
> I'll welcome any comment/suggestion/critic/improvements
>
> Have fun,
>
> Steph
>
> --
> http://nct.ysagoon.com
>
>
> _______________________________________________
> glob2-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/glob2-devel